Introduction to modelling and simulation in textile technology
Abstract:
This chapter introduces the concept of simulation both with and without computers. it then discusses the role of models, including white, black and grey box models. it concludes by considering expert and other knowledge-based systems and their applications.
1.1 Introduction to simulation
Simulation is normally used to analyse systems that are hard or impossible to describe using systems of explicit equations. in particular, this applies to highly complex dynamic systems such as most textile machines and processes. By carrying out a simulation experiment, knowledge about the real system can be gathered comparatively easily in contrast to trials on the machine or during the actual process. This includes practical experiments as well as a simulation using a computer program.
There are several reasons for the application of simulations. investigating the real system can be:
• too time-consuming (e.g. changing the setting of a polymer production plant);
• too costly (e.g. crash trials);
• ethically questionable (e.g. medical trials of a new textile implant);
• too dangerous (e.g. reaction of new textile-based building materials on fire).
Systems that do not yet exist provide another field of application of simulations. this comprises the testing of new plant design concepts or, for example, the development of new aerospace or automotive vehicles using carbon textile reinforced composites. in some cases it is not possible to investigate the actual system directly. A simulation can then be used to gain valuable information which is otherwise unavailable. typical examples are the simulation of the molecular movement in a fluid and the movement of continental plates which takes place over a long time period and thus cannot be observed directly.
1.2 Simulation with and without computers
In general, simulation methods can be divided into those that use a computer and those that do not, as shown in Fig. 1.1. those simulations without computer can be separated into destructive and nondestructive methods. Simulations that use a computer can be based on technical models (e.g. finite element method (FEM), computational fluid dynamics (CFD)), on examples taken from nature (e.g. artificial neural networks, evolutionary algorithms), or on other methods, such as sociological models.
For each simulation, independent of whether it is carried out using a computer or not, it is absolutely crucial that the results can be reproduced every single time that the simulation is run. the use of a simulation must also make sense economically. thus, the model on which the simulation is based is normally simplified as much as possible, but still takes into account the relevant factors that have an influence on the results. This often results in a mathematical model which is valid only for a certain range of the respective values. Hence, it is very important to verify the results of a simulation by running actual trials, e.g. on a machine, in order to ensure that the simulation accurately represents the problem at hand. it must be noted, though, that measurements are never truly accurate, hence the measured results can deviate from reality to a certain extent.
1.2.1 Simulation without a computer
A typical example for a simulation without a computer is a car crash test. In this case the circumstances of an accident are simplified using crash test dummies. these are equipped with sensors that register the mechanical impact the crash would inflict on real persons. This can be achieved by making further simplifications. Experimental investigations of fluid flows also fall into this category when they are carried out, for example, in wind tunnels. Making flows in air visible by using another medium, e.g. water, by applying the theory of similarity, is another typical example for this kind of simulation.
1.2.2 Simulation with a computer
Most technical simulations belong to this kind. Typical examples are calculations of the stresses, e.g. in machine components or textile structures, applying FEM, fluid flow simulations using the CFD method and the simulation of machines and whole production lines, e.g. in a textile mill. Biological simulations such as evolutionary algorithms and neural networks also fall into this category.
1.3 Modelling: white, black and grey box models
Prior to carrying out the actual simulation, a model, in most cases based on equations, must be devised which describes the system with all relevant parameters. In order to find these equations, experiments, often applying experimental design, are normally carried out. Alternatively, the equations are derived from theoretical assumptions. this leads to an image of reality which is either representative of the whole range of parameters or a certain number of parameters. A model is hence an abstract image of the system which is representative of the real system. it is not important that the model exactly mirrors reality in all aspects, but is much more important that the model produces sufficiently accurate results to explain the real case. In general, models are divided into white, black and grey box models as shown in Fig. 1.2.
In order to save time and hence money, most simulation models use a simplified description of the real process or machine. The following methods are commonly used:
• Components or parameters that are not of crucial importance are neglected. in order to determine these components or parameters, experiments applying factorial design can be helpful as the results normally clearly show the size of the effects when changing the settings of these components or parameters.
• Unimportant details are not considered.
• The system is divided into its components which are then investigated separately. However, possible interactions are then hard to determine.
• Combining several attributes into classes can reduce the complexity of a simulation considerably. A major drawback of this approach is that valuable information can also be lost. this method should therefore only be applied carefully and if other methods fail.
1.3.1 White box model
This kind of model is suitable if the inner structure of the system is known. This structure is then deliberately abstracted, modified and reduced to the most important influencing parameters. A typical example is the modelling of a weaving machine (see Chapter 9).
1.3.2 Black box model
If the inner structure of a system is unknown but its behaviour or its interaction with other systems can be observed and modelled, this is called a black box model. A typical example is the use of neural networks to simulate textile processes.
1.3.3 Grey box model
In many cases, only parts of a system are fully known and only a few, but not all interactions between its components are established. the respective model is then called a grey box model. in order to save costs, this approach is widely used. A fuzzy model (see Chapter 4) can be regarded as a grey box model in some cases.
1.4 Expert systems and other knowledge-based models
Many simulation models are based on knowledge of experts of knowledge acquired by running practical trials. in order to design such a model, several steps are required.
• Determination and structuring of explicit and implicit knowledge.
• Build-up of a knowledge database and abstracting of relations between parameters.
• Processing of knowledge, e.g. by inference (see Chapter 4).
1.4.1 Expert systems
This kind of system is based on knowledge of experts which hence allows instructions to be deduced that can be used to solve or to assess certain problems. Expert systems are a branch of artificial intelligence dating back to the 1960s. In many cases, they cannot only be used to make decisions based on existing knowledge but also to give recommendations for new, previously unknown situations. When neural networks are applied, this is called 'learning by discovery'. Figure 1.3 gives an overview of currently used methods.
1.4.2 Case-based systems
These systems use a database in which actual problems and suitable solutions are stored. On a case-by-case basis, the system then determines which stored case is closest to the problem in question. this approach often leads to a viable solution but this method also has some drawbacks: in practice, it is often difficult to implement as similar problems can require totally different solutions so that this approach fails if the database is too small. in order to overcome that difficulty, it is therefore advisable to first weight the parameters describing the problem. Neural networks that use the counterpropagation algorithm are a typical example for the latter approach (see Chapter 2). This can greatly improve the accuracy of the results.
1.4.3 Rule-based systems
These systems comprise sets of rules, e.g. ‘if X then Y’. Based on these rules, conclusions can be drawn to solve the problem at hand. in order to design the rules, normally experts in the field are asked and their 'knowledge' is then fed into the system by formulating rules. A typical example is fuzzy logic (see Chapter 4).
1.4.4 Decision trees
This kind of model is based on a large database which describes the parameter range. It is especially suitable for classification problems. In a first step, those parameters that describe the problem are identified. They are then used to describe the elements to be classified. In a second step, a tree is designed where each branch represents a question that can be answered with 'applies' or 'does not apply'. Following the branches eventually leads to the classification of the unknown object. A typical example for a decision tree is shown in Fig. 1.4. The numbers represent ‘applies’ (1) and ‘does not apply’ (0) which leads to a string of 0 and 1 as the classification result. A typical practical example is the classification of trash particles in cotton. the trash particles can be described according to size, shape, contrast to background, etc. these criteria can then be used to determine the kind of trash particle when an image is taken of the cotton material that contains the trash particles to be analysed.
1.5 Applications of expert systems to textile technology
Expert systems are normally used for highly complex problems when human experts are not readily available or the problem itself is too complex to be solved by them. in these cases, model-based expert systems can represent a successful approach to support the investigator. typical applications of expert systems are:
• Interpretation of data sets (e.g. in digital image processing);
The main drawback of expert systems is their limited knowledge due to their restrictions with regard to the number of available data sets. these in turn can contain errors which does affect the accuracy of the prediction. Besides, the range of values is normally confined to an area in the vicinity of the current problem. this can result in severe problems as only solutions that are close to the known range of parameters may be calculated or considered. this normally does not lead to truly innovative solutions, hence drawing to the conclusion that these systems can never replace the human mind and they are never to be trusted blindly (see example at the end of Chapter 2). A typical example for the disastrous consequences of blindly trusting expert systems is the Black Monday in 1987 when unsupervised computer programs caused havoc in the stock market when cascades of wrong decisions that were programmed into the systems led to a complete loss of control by traders.
1.6 References and further reading
There is a wide range of books available that cover different aspects of simulation. Some deal with general topics, others concentrate on certain methods. This subsection briefly describes some recently published books that are recommended reading. this is not intended to represent a complete list but only a selection of books.
Sokolowski and Banks (2009) explains the principles of modelling and simulation for both discrete and continuous systems. the authors of the book cover a wide range of applications not only in engineering but also in computer science and related fields. Practical examples from transportation, medicine and business decision illustrate the concepts of simulation and modelling presented in this book. Hence, it gives a good introduction into the subject although it does not go into too many details regarding the algorithms behind the examples.
Tettamanzi and Tomassini (2010) was written as a workbook for students and gives a comprehensive description of neural networks, fuzzy systems and evolutionary algorithms. It also details the combination of these systems, e.g. evolutionary design of neural networks, neuro-fuzzy systems. It is therefore especially suited for those interested in understanding the basics of soft computing. A concise description of computational fluid dynamics is given in Wendt (2010). The book covers the basics as well as a wide range of applications and is written by experts in the field and hence a good introduction to the subject.
Kleijnen (2010) is an excellent introduction into many methods that can be used to verify simulation results. the focus is laid on design and analysis of simulation experiments. Polynomial regression models as well as simulation optimization and screening design methods are covered and explained with many examples. An excellent book on the design of experiments (DoE) is Antony (2003). Its examples cover a wide range of manufacturing problems and the book is hence especially suitable for engineers. Another book which is designed to serve as a concise workbook with examples covering 2, 3 and up to 31 factor experiments is Barrentine (1999).
A very useful book on the simulation of textiles is Majumdar (2011a). It describes various soft computing techniques from yarn manufacturing through fabric production to textile properties, applications and quality evaluation. Another recent publication by the same author cites more than 300 useful references relating to textile applications of soft computing (Majumdar, 2011b). Chen (2010) presents a wide range of mathematical models, e.g. for fibres, yarns, fabrics and composite structures. Many examples are explained in great detail which gives an excellent overview of the field. Both are therefore perfect complements to this book which covers many of the respective machines and processes. Another book covering different aspects of textile production and the possibilities to simulate the respective processes is Sztandera and Pastore (2010). It outlines the basics of soft computing and details a large number of applications in the textile industry.
Antony, J. Design of experiments for engineers and scientists. Butterworth-Heinemann; 2003.
Barentine, L.B. An introduction to design of experiments: A simplified approach. American Society for Quality; 1999.
Chen, X. Ed.Modelling and predicting textile behaviour. Woodhead Publishing, 2010.
Kleijnen, J.P.C. Design and analysis of simulation experiments. Springer; 2010.
Majumdar A., ed. Soft computing in textile engineering. Woodhead Publishing, 2011.
Majumdar, A. Soft computing in fibrous materials engineering. CRC Press Inc; 2011.
Sokolowski J.A., Banks C.M., eds. Principles of modeling and simulation: A multidisciplinary approach. Wiley, 2009.
Sztandera L.M., Pastore C., eds. Soft computing in textile sciences (studies in fuzzyness and soft computing). Physica-Verlag, 2010.
Tettamanzi, M.T., Tomassini, M. Soft computing: Integrating evolutionary, neural and fuzzy systems. Springer; 2010.
Wendt J.F., ed. Computational fluid dynamics: An introduction. Springer, 2010.